[musicreo]

I have done some new measurements and now also used my AKG701. This time the channel balance was necessary and helped a lot to obtain again a very nice result. The AKG701 and the Sennheiser HD555 really sound very similar and both now deliver a realistic room image.

The most positive effect for me in this new measurment was to put one of the speakers to the center position. This center sounds much better than the ones recorded from the left or right speaker position (probably due the different reflections of the walls as the center speaker does now have 1.38 m more space to both sides.)

 

[johnn29]

I'm going to investigate these errors soon. There's something going on for sure.

Pre is the binaural recording without any processing, room is the room recording as is and post is the binaural measurement with all the processing (room correction, channel balancing, headphone compensation,... ). I'll add these to the instructions.


I think the biggest value of head tracking would come from added comfort because you can safely mice your head around without the fear of that imaging gets ruined. That being said I have to confess that I've never tried head tracking...

Can you change the axis for the room correction report so the Y axis scales to show the full frequency range better? Currently because of the huge negative gain from 0-20db you can't really see the plots in much detail and they all look quite flat. Would love to see those in proper detail.

What I find puzzling though is the decay time. The pre corrected measurement has a decay time in the order of seconds. See the attached images for pre and post. How is Impulcifer correcting that?

After being an AVR user with all the different room correction solutions, ultimately coming to the conclusion that room correction over the transition frequency is pointless due to Toole's work - it's kind of amazing to get room correction at an ideal point and then have that point move around with your head. No averaging, no multi point measurements.

 

[jaakkopasanen]

Can you change the axis for the room correction report so the Y axis scales to show the full frequency range better? Currently because of the huge negative gain from 0-20db you can't really see the plots in much detail and they all look quite flat. Would love to see those in proper detail.

What I find puzzling though is the decay time. The pre corrected measurement has a decay time in the order of seconds. See the attached images for pre and post. How is Impulcifer correcting that?

After being an AVR user with all the different room correction solutions, ultimately coming to the conclusion that room correction over the transition frequency is pointless due to Toole's work - it's kind of amazing to get room correction at an ideal point and then have that point move around with your head. No averaging, no multi point measurements.

I'll see what I can do with the FR plot Y-axis. The limits are set automatically right now.

Decay in the decay graph starts at around 3500ms and reaches noise floor at around 4000ms so it's only half a second. Everything before the largest spike in the decay graph are the harmonic components. Impulcifer crops out everything before the left-most peak and also everything after the noise floor has been reached. This is why there is the drop in the post waterfall graph and why you can't see the impulse responses of the harmonic components there.

 

[jaakkopasanen]

Channel balance of the demo recording has been bugging me even with channel balancing using trend method. Giving more gain to left ear doesn't really do the trick and it creates this uncanny uneven pressure feeling so it's not the way to go about it. Now I tried adjusting the channel balance of the speakers before HeSuVi and that worked a lot better. I added this to config.txt:

Copy: L=1.078*L R=0.928*R
Include: HeSuVi\hesuvi.txt

That's 1.3 dB difference in favor of FL. HeSuVi's speaker placement adjustment doesn't do this and I don't really know what it does.

It's possible that the level matching in room correction is not working correctly. I will have to investigate what are the speaker levels in room recordings, level matched room recordings and the in the final HRIR output. It would be really weird if there would not be a level difference when manually adjusting for that fixes the channel balance issue.

 

[johnn29]

Trying to nail a near field measurement this morning. It's a bit of a pain to move everything and extend the wires but hoping it's worth it! The external webcam really helped with the UMIK placement.

 

[Joe Bloggs]

Channel balance of the demo recording has been bugging me even with channel balancing using trend method. Giving more gain to left ear doesn't really do the trick and it creates this uncanny uneven pressure feeling so it's not the way to go about it. Now I tried adjusting the channel balance of the speakers before HeSuVi and that worked a lot better. I added this to config.txt:

Copy: L=1.078*L R=0.928*R
Include: HeSuVi\hesuvi.txt

That's 1.3 dB difference in favor of FL. HeSuVi's speaker placement adjustment doesn't do this and I don't really know what it does.

It's possible that the level matching in room correction is not working correctly. I will have to investigate what are the speaker levels in room recordings, level matched room recordings and the in the final HRIR output. It would be really weird if there would not be a level difference when manually adjusting for that fixes the channel balance issue.

Are you doing anything about time alignment at all?

 

[jaakkopasanen]

Are you doing anything about time alignment at all?

Yes, I am. Each speaker delay is adjusted so that the distances from the speakers are equal to the center of the head. Interaural time difference is not touched but relies on the measurement itself. The speaker delay adjustment is done on the basis of the primary ear (left ear for left side speakers, right for right).

I did take a quick look at the demo recording FR levels and there doesn't seem to be much of a difference. So I'm puzzled.

 

[Joe Bloggs]

Yes, I am. Each speaker delay is adjusted so that the distances from the speakers are equal to the center of the head. Interaural time difference is not touched but relies on the measurement itself. The speaker delay adjustment is done on the basis of the primary ear (left ear for left side speakers, right for right).

I did take a quick look at the demo recording FR levels and there doesn't seem to be much of a difference. So I'm puzzled.

Do you want to send me the IR files before and after processing to have a look?

 

[jaakkopasanen]

Do you want to send me the IR files before and after processing to have a look?

Do you have Impulcifer installed? The recordings can be found in the data/demo folder. When running Impulcifer pointed to that folder it will produce responses.wav which contains the unprocessed IRs and hrir.wav which contains the processed IRs.

 

[johnn29]

Room correction on the nearfield HRIR I produced yesterday has a much bigger impact. I suspect it's because I'm really only picking up on bass improvements. My measurement at my regular seating position used my sub, the nearfield one was without a sub or any AVR bass management/correction. The virtual room correction takes care of that quite nicely.

I did get an unusual situation when moving about 1.5m away from the speakers though - the HRIR produced odd effects where the virtual channels bled into each other. For example on a channel ID check I could hear a very quiet echo from the rear left/rear right when it was the front channels being ID'd. I suspected it was the room correction bootsing bass frequencies to the directional level and that was correct. I'll raise an issue in Github. Not an issue for me really - but worth fixing.

 

[jaakkopasanen]

So today I don't really need the -1.5 dB on FL speaker channel to get good channel balance with demo recording. Maybe I'm going crazy. Or maybe it's about how I happen to put headphones on my head.

 

[johnn29]

In my experience it's so hard to do perceptual testing with HRIR's. One day you feel like you hear one thing, and another day it's not there. My perception even gets confused depending on where I'm sitting.

With headphone placement I suspect the best headphones are those with active EQ's that compensate for placement on your head so they always deliver a consistent frequency response. But those tend to have loudness curves baked in, which aren't ideal.

 

[jaakkopasanen]

In my experience it's so hard to do perceptual testing with HRIR's. One day you feel like you hear one thing, and another day it's not there. My perception even gets confused depending on where I'm sitting.

With headphone placement I suspect the best headphones are those with active EQ's that compensate for placement on your head so they always deliver a consistent frequency response. But those tend to have loudness curves baked in, which aren't ideal.

I've noticed this myself too. Also a note of interest here is that I've never had reliable and perfect channel balance with any of my speaker systems, they always tend to lean towards right side (unless I compensate for it), just like these demo recordings. So maybe it has something to do with asymmetrical head and ear shape which isn't problem with sounds that originate from the actual physical source but isn't ideal with phantom center.

@Joe Bloggs You can also find the impulse responses now in the speaker balance graphs experiment folder without running Impulcifer here: https://github.com/jaakkopasanen/Impulcifer/tree/master/research/speaker-balance-graphs

 

[johnn29]

This could be of relevance? https://www.audioholics.com/room-ac...ons-human-adaptation/what-do-listeners-prefer

A possible factor in the acceptance of some reflected energy may be related to the interaural crosstalk that audibly distorts the phantom center image in stereo. When in the stereo seat, in the absence of reflections there is a huge 2 kHz dip - it is not hard to hear when one is in a dominantly direct sound field. As shown in Figure 9.7 (below) it is somewhat alleviated by reflected sounds - even improving speech intelligibility. Remember that normally this is the sound of the featured artist. There can be no better argument for a center channel.

I've never found phanton centers to be convincing. Headphones is the closest I've gotten to it being bang on with synthetic HRTFs and some of the ooyh presets and now Impulcifer. When I listen to my real system I always use Pro Logic to mix it to surround. I don't really like the rear channels much, but it does a great job of extracting the center image and that's better for multiple positions/seats. I've tried ffdshow audio and AC3 filter to do the same thing with stereo sources for HRTF. It works to an extent - but the Dolby Surround processor isn't as good as on my AVR.

 

[jaakkopasanen]

I had an idea that perhaps the headphone compensation isn't working as intended because it's only trying to equalize the headphone frequency response flat at the microphone and is therefore only a feed-forward compensation. I tested this hypothesis by implementing a feedback compensation which does the headphone measurement using sine sweep sequence convolved with the HRIR measurement. This way the frequency response of the heaphone measurement should be the same as the frequency response of the HRIR measurement. The cool thing about this idea is that one doesn't need to know the feed-forward target of the headphones, it can be whatever as long as the virtualized speakers have the same frequency response as the physical speakers.

Hypothesis was proven to be incorrect. The headphone compensation for both feed-forward and feedback measurements is almost exactly the same with only significant difference in sub 30 Hz region. This also proves that headphones really need to be flat at the microphone as was previously assumed. Check out the experiment and results here: https://github.com/jaakkopasanen/Impulcifer/tree/master/research/headphone-feedback-compensation

My mic calibration shows the binaural mics to have a level difference of about 1 dB but now both headphone compensation methods indicate 3 dB level difference. Channel balancing in impulcifer reduces the right side volume so I would suspect that the level difference according to the mic calibration is correct. I find it unlikely that the level difference would be caused by the headphones either. Very strange...